AN AUTOREGRESSIVE BASED LFM REVERBERATION SUPPRESSION FOR RADAR AND SONAR APPLICATIONS MrPMohan Krishna 1, AJhansi Lakshmi 2, GAnusha 3, BYamuna 4, ASudha Rani 5 1 Asst Professor, 2,3,4,5 Student, Dept of ECE, Sri Sivani College of Engineering, JNTUK, AP, (India) ABSTACT In a real time noisy environment scenario, improving the target detection performance is one of the most important issues when active sonar and radar systems are used for target detection In this paper, we are aiming to suppress reverberation that is present while transmitting a Linear Frequency Modulated (LFM) signal The motivation behind this paper is to convert the LFM reverberation to whiten using parameter based Autoregressive (AR) model As LFM reverberation is a frequency dependent which is severely colour noise, so we are focused to attain stationary frequency and the adjacent signal block is used as a reference signal of proposed technique The study of performance comparison of parametric based AR prewhitening using Levinson-Durbin and Burg methods were studied with respect to LFM reverberation suppression and results are presented using MATLAB tool Keywords: Linear Frequency Modulated Signal, Reverberation and Autoregressive I INTRODUCTION Noise is an uncontrollable component as it is frequency independent Reverberation is a self generated controllable noise which is frequency dependent While detecting a low speed targets, Reverberation and Cluttering deeply affects the target detection performance of active sonar and radar systems respectively If Reverberation exists, it is difficult to distinguish the actual and fault targets As Reverberation is severely colored, correlated strongly with the emitted signal and non stationary in both frequency and time domain, matched filter fails to give the optimized output Reverberation or cluttering cannot be eliminated completely but can be suppressed using different signal processing techniques Prewhitening technique is mainly used to whiten the reverberation before applying to matched filter In active sonar and radar systems, the term like autoregressive (AR) have been used for modeling processes like reverberation suppression similar to all-pole filtering mode 12 LFM SIGNAL In the radar literature, LFM is known to be easily generated by a variety of technology and has a superior performance in pulse compression Pulse compression is used to increase range resolution and signal to noise ratio To transmit a long pulse that has a bandwidth corresponding to a short pulse, pulse compression technique is required Linear frequency modulation (LFM or chirp) signals are widely used in 436 P a g e
information systems Linear frequency modulation is a technique used to increase the waveform bandwidth while maintaining pulse durations such that τ>>1/β =>Time-bandwidth product βτ >>1 A linear frequency modulated signal is defined as 2 Ɩ fm ( t ) = e j2π(f t+mt 2 ) ; for 0 t T = 0 ; otherwise The instantaneous frequency of the LFM signals is given by ƒ i (t) = d(ƒ 0 t +( BW / 2T)t 2 ) / dt = ƒ 0 + (BW / T) II ADVANTAGES OF LFM SIGNAL Increasing the duration of transmitted pulse increases its energy and improves target detection capability; conversely, reducing the duration of a pulse improves the range resolution of the radar Pulse compression techniques enable to decouple the duration of the pulse from its energy by effectively creating different durations for the transmitted pulse and processed echo III AR MODEL In AR model the future values are estimated based on the weight sum of past valuesa time-varying autoregressive (TVAR) approach is used for modelling non-stationary signals, and frequency information is then extracted from the TVAR parameters An auto regressive model is also known in the filter as an infinite impulse response filter or an all pole filter Xt =Σ ai xi- i + et where ai-autoregressive coefficients Xt-series under investigation N-order of the filter All zero model (in statistics, moving average (MA)model) Mixed pole-zero model is called auto regressive moving average (ARMA) model) 31 Prewhitening Processing 437 P a g e
Step 1To generate transmitted signal using linear frequency modulated signal with low and high speed conditions Step 2 Divide the received signal into blocks using a sliding window Step 3 Whiten the current signal block with AR prewhitening in time domain Step 4 Compare Levinson and burg method results 32 Yule-Walker Equations The linear relationship between XX (m) and the {a K } parameters these equations called the yule- walker equations can be expressed in the matrix form γᵪᵪ 0 γᵪᵪ 1 γᵪᵪ 2 γᵪᵪ p γᵪᵪ 1 γᵪᵪ 0 γᵪᵪ 1 γᵪᵪ( p + 1) γᵪᵪ p γᵪᵪ p 1 γᵪᵪ p 2 γᵪᵪ(0) This matrix is symmetric, but is also toeplitz matrix, which means the inverse can be performed efficiently, using an iterative algorithm 33 Levinson_Durbin Algorithm In p th order forward linear predictor, the current sample of speech is predicted from a linear combination of th p-past samples Levinson algorithm is achieved by the minimizing the MSE of the forward prediction error and then applying the toeplitz feature of the auto correlation matrix,a fast algorithm based on order recursion is derived by levinsonsource code can be found in the software packet, because of this algorithm is based on the solution of toeplitz autocorrelation matrix it is also called autocorrelation methodthe levinson durbin algorithm is computationally efficient algorithm for solving the normal equations for prediction coefficients 34 Burg Algorithm Autoregressive (AR) modelling is used for analyzing stationary stochastic processes for many different applications, eg, radar geophysics, and economics A comparison of various estimators of AR-parameters showed that the Burg algorithm is the preferred estimator for AR-parameters The Yule Walker algorithm can be severely biased After a large true reflection coefficient, estimates for higher order reflection coefficients suffer from a bias of order 1 instead of the smaller bias of order 1/N that is present in other estimation methods The least squares estimator and the forward-backward least squares estimator have a greater variance than Burg In addition, they may yield unstable models In many applications, the duration of one uninterrupted measurement is submitted to practical limitations However, it is often possible to obtain several separate segments of data The Burg algorithm 1 a1 ap = σ 2ᵤ 0 0 438 P a g e
IV RESULTS AND TABLES Levinson_Durbin results 1 ( a ) 1( b ) 1 (c) 1 (d) 1(e) 1 (f) Fig1: Target detection results at p=10 & SNR=20db at Levinson_Durbin 1( a )before prewhitening in time domain 1 ( b ) before prewhitening in mesh plot 1 ( c ) after prewhitening in time domain 1(d) after prewhitening in mesh plot 1( e ) matched filter output before prewhitening 1( f ) matched filter output after prewhitening BURG RESULTS 439 P a g e
2 (a) 2(b) 2(c) 2(d) 2 (e) 2 (f) Fig2: Target detection results at p=10 & SNR=20db at burg 2 (a) before prewhitening in time domain 2 (b) before prewhitening in mesh plot 2(c) after prewhitening in time domain 2 (d) after prewhitening in mesh plots 2 (e) matched filter output before prewhitening 2 (f) matched filter output after prewhitening Figure(1) shows the target detection results using levinson_durbin methods at filter order(p) =10 and SNR =20 dbfig1(a)&1(b) shows target detection results before prewhitening at low speed targets at t =14 and amplitude is 1932 10⁷at high speed target s t =2892 and amplitude is 4338 10⁷figure 1(c)&(d) shows target detection results after prewhitening at low speed targets at t = 14 and amplitude is 1928 10, at high speed targets at t = 2892 and amplitude is 1928 10⁷fig 1(e)&1(f) are matched filter outputs before and after prewhitening 440 P a g e
Figure((2) shows the target detection results using burg methods at filter order(p) =10 and SNR =20 db fig2(a)&2(b) shows target detection results before prewhitening at low speed targets at t =14 and amplitude is 2225 10⁷ at high speed target s t =2892 and amplitude is 499 10⁷ figure 2(c)&(d) shows target detection results after prewhitening at low speed targets at t = 14 and amplitude is 293 10⁷, at high speed targets at t = 2892 and amplitude is 6645 10⁷ fig 2(e)&1(f) are matched filter outputs before and after prewhitening 1 COMPARING THE RESULTS BEFORE & AFTER PREWHITENING BY USING LEVINSON_DURBIN AT DIFFERENT SNR LEVELS AMPLITUDE SNR LEVINSON-DURBIN METHOD BEFORE PREWHITENING AFTER PREWHITENING Low speed target at t=14 High speed target at t=2892 Low speed target at t=14 High speed target at t=2892 20 db 15 db 10 db 5 db 0 db -5 db -10 db -15 db -20 db 1932 10⁷ 2189 10⁷ 1988 10⁷ 2007 10⁷ 1925 10⁷ 2022 10⁷ 2038 10⁷ 2007 10⁷ 2016 10⁷ 4338 10⁷ 4812 10⁷ 472 10⁷ 4782 10⁷ 491 10⁷ 4924 10⁷ 4931 10⁷ 4919 10⁷ 4917 10⁷ 1928 10⁷ 2301 10⁷ 1983 10⁷ 2042 10⁷ 1975 10⁷ 2014 10⁷ 2032 10⁷ 2123 10⁷ 2043 10⁷ 4358 10⁷ 5043 10⁷ 4615 10⁷ 5282 10⁷ 5632 10⁷ 4915 10⁷ 4906 10⁷ 521 10⁷ 4968 10⁷ 2 COMPARING THE RESULTS OF BEFORE & AFTER PREWHITENING BY USING BURG AT DIFFERENT SNR LEVELS SNR 20 db 15 db 10 db 5 db 0 db -5 db -10 db -15 db -20 db AMPLITUDE BURG METHOD BEFORE PREWHITENING AFTER PREWHITENING Low speed target at t = 14 High speed target at t = 2892 Low speed target at t = 14 2225 10⁷ 499 10⁷ 293 10⁷ 2189 10⁷ 4812 10⁷ 2422 10⁷ 2074 10⁷ 494 10⁷ 2669 10⁷ 2069 10⁷ 4883 10⁷ 2278 10⁷ 2006 10⁷ 485 10⁷ 2003 10⁷ 2013 10⁷ 487 10⁷ 184 10⁷ 1996 10⁷ 4919 10⁷ 2273 10⁷ 2013 10⁷ 4915 10⁷ 1813 10⁷ 2016 10⁷ 4918 10⁷ 1982 10⁷ High speed target at t =2892 6645 10⁷ 5291 10⁷ 6575 10⁷ 4364 10⁷ 4827 10⁷ 4445 10⁷ 5623 10⁷ 4422 10⁷ 4902 10⁷ 441 P a g e
IV CONCLUSION In this paper the study of performance of the AR prewhitening method for LFM reverberation suppression has been explored The simulation results are analyzed by performing AR prewhitening on the received signal data Simulation result analysis with received signal data proves that the proposed method is levinson-durbin method provide an improved whitening performance In future, moving average (MA) and autoregressive moving average (ARMA) methods would be used to suppress reverberation REFERENCES [1] J G Proakis, D G Manolakis: Digital Signal Processing: Principles, Algorithms, and Applications, PrenticebHall, 1996, 3rd edition [2] J G Proakis, D G Manolakis: Digital Signal Process in Principles, Algorithms, and Applications, Prentice Hall, 2007, 4th edition [3] AV Oppenheim, R W Schafer: Discrete-time signalprocessing, Prentice Hall, 1999, 2nd edition [4] Sudhir Kumar and Avinash Kumar Dubey, Parameter Estimation of Frequency Modulated Continuous Wave (FMCW) Radar Signal using Wigner-Ville Distribution and Radon Transform, International Journal of Advanced Research in Computer and Communication Engineering Vol 3, Issue 6, June 2014 [5] Ruhang Wang, Jianguo Huang, Tian Ma, and Qunfei Zhang, Improved space time prewhitener for linear frequency modulation reverberation using fractional Fourier transform, Acoustical Society of America, J Acoust Soc Am 128 (6), December 2010 442 P a g e